#two-stage-detectors

Question 1: Single choice

What is the main structural difference between single-shot multibox detection (SSD) and Faster R-CNN?

SSD builds on a pretrained CNN backbone, while Faster R-CNN has to train its whole CNN from scratch1
SSD relies on selective search for its region proposals, while Faster R-CNN learns them with a proposal network2
SSD predicts from dense anchors in one pass; Faster R-CNN first proposes regions, then classifies each3
SSD predicts a pixel mask for every object, while Faster R-CNN predicts only the bounding boxes4